Testing Wagner versus Keynesian Hypothesis for Pakistan: The Role of Aggregate and Disaggregate Expenditure

Transcription

1 MPRA Munich Personal RePEc Archive Testing Wagner versus Keynesian Hypothesis for Pakistan: The Role of Aggregate and Disaggregate Expenditure Wajid Ali and Kashif Munir University of Central Punjab 9 September 2016 Online at MPRA Paper No , posted 16 October :27 UTC

2 Testing Wagner versus Keynesian Hypothesis for Pakistan: The Role of Aggregate and Disaggregate Expenditure Wajid Ali * & Kashif Munir University of Central Punjab, Lahore, Pakistan * Department of Economics, University of Central Punjab, Lahore, Pakistan Associate Professor, Department of Economics, University of Central Punjab, Lahore, Pakistan. Phone: , Fax: , kashif.munir@ucp.edu.pk, kashif_ok@hotmail.com

3 Abstract The objective of this study is to examine the relationship between government expenditure (at aggregate as well as disaggregate level) and economic growth for Pakistan. The study further aims to find causal relationship for existence of applicability of Wagner s or Keynesian hypothesis. Engle and Granger (1987) two step procedure and Granger causality test (1969) are employed for time series data for the period 1976 to Results suggest that only expenditure on social, economic and education services have proposed long-run association with economic growth in five of basic versions of Wagner s law for Pakistan. The causality tests are showing mix results regarding existence of applicability of Wagner s or Keynesian hypothesis. Expenditure on current subsidies, expenditure on defence, current expenditure and developmental expenditure are in favor of Wagner s law in most of the cases, where causality runs from economic growth to government expenditure. Results of expenditure on social, economic and education services are in line with existence of Keynesian hypothesis, where causality flows from government expenditure to economic growth. On the basis of results, one may conclude that government should invest for expenditure on social, economic and education services to achieve sustainable economic growth by spending more on human resource development. Keywords: Government Expenditure, Economics Growth, Cointegration, Causality, Pakistan. JEL: C32, E62, H51, H52 1

4 1. Introduction The relationship between public spending and national income has been an important subject of analysis and discussion for decades among economists (Peacock & Wiseman, 1961; Gupta, 1967; Oktayer, 2013). Government attempts to stimulate economic growth through various policy instruments. One of the important instrument of fiscal policy is public spending which is used to influence economic growth. (Lahirushan & Gunasekara, 2015). The government expenditure policy has a crucial role to operate functions of the economy whether it is developed or underdeveloped. The recessionary/expansionary periods reduce/increase the abilities of state to enhance economy through fiscal policy instruments unless share of public spending to GDP increases/reduces (Magazzino, 2012). Government revenue was more important than government expenditure till 20 th century, while functions and activities of state were confined to a specific limit (World Development Report, 1988). The thinking about functions of state has been changed and present state is now considered to be a welfare state for the economy. The state increases welfare of nation through spending on developmental projects and also social, economic and education services i.e. employment, health, agricultural and industrial development, fresh and clean water (World Development Report, 1988). The rapid economic growth is un-desirable without state influence, while private institutes are only curious about to earn profit and to survive in economy. There exists mainly two approaches in the literature regarding pubic spending and growth i.e. Wagner s law or Keynesian hypothesis. The fundamental arguments for these two approaches is rely on causal link between public spending and growth (Samudram, 2009). Wagner (1883) states public spending and growth have positive association (Henrkeson, 1993). He is of the view that during industrialization process, government expenditure tends to expand in case of increase in per capita income of a nation, which indicates causality flows from output to public spending (Cheltsos & Kollias, 1997). While, Keynes (1936) postulates government expenditure is exogenous policy instrument which is used to accelerate growth and to correct short-run as well as long-run cyclical fluctuations (Ansari et al., 1997). The public spending is not cause of economic growth and spending do not play decisive role to accomplish growth, so that causality flows from public spending to national output. 2

5 The relationship between public spending and economic growth is one of the most debated issue, because government expenditure policy is important tool in economic analysis (Montiel, 2010). Wagner s and Keynesian hypothesis have been empirically tested for both developed and developing countries (Ansari et al., 1997; Biswal et al., 1999; Islam, 2001; Samudram et al., 2009). The literature in favor of Wagner s or Keynesian hypothesis is divided into three strands. First strand is based on validity of Wagner s hypothesis, where causality is unidirectional flows from economic growth to public spending (Ansari et al., 1997; Islam, 2001; Chow et al., 2002; Faris, 2002; Monteil, 2010; Abdullah & Mamoor, 2010; Kumar et al., 2012; Barra, 2015; Thabane & Lebina, 2016). Second strand depends upon applicability of Keynesian hypotheses, where causality is also unidirectional flows from public spending to economic growth (Halicioglu, 2003; Babatunde, 2011). Third strand is based on existence of both Wagner s and Keynesian hypothesis, where causality is bidirectional for public spending and economic growth (Biswal et al., 1999; Dritsakis & Adamopoulos, 2004; Ziramba, 2008; Katrakilidis & Tsaliki, 2009; Samudram, 2009; Ono, 2014). In sum, the literature provides existence of positive or negative relationship between public spending and economic growth for developed and underdeveloped countries and also reveals applicability of Wagner s or Keynesian hypothesis. However, there is limited work found for existence of relationship between government expenditure (at aggregate as well as disaggregated level) and economic growth specifically for Pakistan. There is less discussion for Pakistan on six basic versions of Wagner s law developed by Peacock-Wiseman (1961), Pryor (1968), Goffman (1968), Michos (1975), Mann (1980) and Murthy (1993) respectively. The present study aims to fill this gap by analyzing relationship between total public spending, its components and economic growth by employing six of the mathematical formulations of Wagner s law. The objective of this study is to examine relationship between government expenditure (at aggregate as well as disaggregate level) and economic growth for Pakistan from 1976 to The specific objectives of the study are as follow: to analyze relationship between government expenditure and economic growth, and to test validity of Wagner s or Keynesian hypothesis. The study contributes to existing literature by investigating relationship between government expenditure (at aggregate as well as disaggregate level) and economic growth for Pakistan. The applicability of Wagner s or Keynesian hypothesis is tested by incorporating six of mathematical formulations of Wagner s law for direction of causality between variables. The findings of this 3

6 study will help policy makers and government to design appropriate policies to accelerate the pace of growth. Theoretical model is based on Peacock-Wiseman (1961), Pryor (1968), Goffman (1968), Michos (1975), Mann (1980) & Murthy (1993) versions of Wagner s hypothesis. Whereas, total government expenditure at aggregate level as well as disaggregate level is treated as dependent variable, while GDP is taken as explanatory variable. Time series data for Pakistan over the period 1976 to 2015 has been utilized for analysis. This study uses time series data so Augmented Dickey Fuller (ADF) and Phillips-Perron (PP) unit root tests are applied to check non-stationary properties of time series, because the series having unit root will mislead results. Engle and Granger (1987) methodology of cointegrtaion is used to examine long-run relationship between variables of analysis. The causality tests allow us to discuss the causal links between variables, therefore Granger causality test is applied to determine the existence of validity of Wagner s or Keynesian hypothesis. The rest of study is organized in following manners. Literature review on relationship between public spending and economic growth is discussed in section 2. Model, methodology and data are explained in section 3. Section 4 presents empirical results of the study. Section 5 consists of concluding remarks and policy recommendations. 2. Literature Review Government expenditure policy plays a pivotal role to operate the economy whether it is developed or underdeveloped, because it helps to enhance economic activities of these nations. The role of government towards economic growth is important theoretically and empirically. Ansari et al. (1997) analyzed hypothesis of income and expenditure for three African countries such as South Africa, Ghana and Kenya. They applied Holmes-Hutton statistical procedure for time span ranging from 1957 to The results showed no cointegration between public spending and output for Ghana, Kenya and South Africa. The causality results suggested causality flows from economic growth to spending in favor of Wagner s law for Ghana. There was no causality for Kenya. There was unidirectional causality in favor of Keynesian hypothesis for South Africa. They concluded that many factors i.e. defence expenditure do not helpful to increase growth, but developmental expenditure on roads, dams, bridges, transportation and communication enhance growth and development in economy. Biswal et al. (1999) found relationship between total public spending, its various subcomponents and national income for 4

7 Canada. They applied cointegration methodology and causality tests for time span ranging from 1950 to The results showed absence of cointegration between components of public spending except current outlays and total current expenditure. These two expenditure expressed bi-directional and others had unidirectional causality in favor of Wagner s and Keynesian hypothesis. They concluded that stabilization policy cannot use transfer payments as their instrument but these could be helpful in implementing the fiscal austerity programme. Spending on other goods and services or public debt could offer significant stabilization power. Finally, current outlay, public purchases, spending on wages and salaries are not good choice for policy recommendations. Burney (2002) examined association between public spending and socioeconomic variables, along with level of income for Kuwait. Cointegration methodology was employed for time span ranging from 1969 to Results found little support for cointegration between variables under analysis. Causality results were not in favor of Wagner s hypothesis. He concluded that GNP, GDR, trade openness, revenues supply, population configuration and revenue constraints may influence long-run government expenditure. Wagner s hypothesis may not be valid for Kuwait, because large portion of its revenue is assessed exogenously. While for those countries where Wagner s hypothesis is applicable, their income is determined within domestic economy. Faris (2002) investigated relationship between public spending and output for case of Gulf Cooperation Council countries. Cointegration and causality test were employed for time span 1970 to Results described cointegration between real total government spending, its components and economic growth such as capital and current outlay. The causality test explained that Wagner s hypothesis is applicable in GCC countries except Bahrain. In Bahrain, both Wagner and Keynesian hypothesis exist. He concluded that Keynesian hypothesis is not applicable among these countries, because a huge part of government spending is devoted to current expenditures such as discount on utility bills, recreational facilities and defence spending. Even these expenditure may not helpful in growth enhancing actions. Therefore, efficiency and effectiveness of expenditure is prerequisite to gain from government expenditure. Halicioglu (2003) empirically examined long run association between public spending and output for Turkey. Cointegration and causality test are applied for time span ranging from 1960 to Results suggested that there exists cointegration between public spending and economic growth. The results of causality suggested neither unidirectional nor bidirectional causality in 5

8 favor of Wagner s vs. Keynesian hypothesis. Iyare et al. (2004) analyzed long-run association between public spending and national income for nine Caribbean countries. Cointegration and causality tests were employed for the period 1950 to Results indicated absence of cointegration between public spending and output for Caribbean countries except Grenada, Guyana and Jamaica. Results of causality test were mixed, but in most of the countries, Wagner s hypothesis is applicable. Chang et al. (2004) empirically estimated the applicability of Wagner s hypothesis on ten countries including seven industrialized and three emerging industrialized countries of Asia. They used Johansen and Juselius cointegration (1992) for the period 1951 to Cointegration results suggested public spending and output is moving together in long-run for Japan, South Korea, Taiwan, UK and US. Results of causality test discussed that Wagner s hypothesis is applicable for Japan, US, UK, South Korea and Taiwan in favor of Wagner s law. For remaining five countries, neither Wagner s nor Keynesian hypothesis validated. Dritsakis et al. (2004) examined association between public spending and output for Greek. Cointegration and causality tests applied for period 1960 to Results dictated presence of cointegrtion between variables. Results of long-run causality test suggested unidirectional causality runs from output to government expenditure for major parts of government expenditure i.e. education, healthcare and culture. The short-run causality test revealed Keynesian theory is applicable with general government spending, while Wagner s hypothesis is validated for aggregate expenditure and social welfare expenditure. They concluded that both Wagner s and Keynesian hypothesis are applicable for most of expenditure in Greek. This implies that government expenditure and economic growth increase each other at the same time. They further elaborated that Wagner s hypothesis is valid for total expenditure and social welfare expenditure. While the causality results for general government expenditure do favor Keynesian hypothesis. Loizides and Vamvoukas (2005) proposed link between public spending, government size and output for UK, Ireland and Greece. Cointegration methodology was used for bivariate as well as trivariate analysis and causality test for the period 1960 to Bivariate analysis were not in favor of long-run relationship. As inflation and unemployment is included, the trivariate analysis showed the presence of long-run association. The results of causality suggested that causal link flows from public spending to economic growth in favor of Keynesian hypothesis in short run for all countries and in long run for UK and Ireland. Wagner s hypothesis is applicable for 6

9 Greece. When inflation is included, causality flows from economic growth to government size for UK, which is also indicating existence of Wagner s hypothesis. They concluded that impact of growth on public sector is significant, which indicates government expenditure helps to promote growth. Rehman et al. (2007) examined association between government expenditure and output for Pakistan. Johansen co-integration and causality tests were applied for the period 1972 to Results elaborated the existence of cointegration between public spending ratio, real per capita GDP, financial development and trade openness. They concluded that major determinants of public spending consist of per capita output, financial developments and trade openness. These determinants have long-run and significant impact, therefore policy makers can take different actions to boost economy with the help of these determinants. Ziramba (2008) analyzed relationships between real public spending and real output for South Africa. ARDL bound test was applied for the period 1960 to Results suggested that public spending and output are moving together in long-run. Results of short-run causality indicated both Wagner s and Keynesian hypothesis are valid, while in long-run causality do not follow any direction. Katrakilidis and Tsaliki (2009) analyzed association between public expenditure and output for Greek economy. ADRL co-integration approach was employed for the period 1968 to Results suggested presence of cointegration between variables. Results of causality analysis suggested a bidirectional causal relationship which support both Wagner s and Keynesian hypothesis. They concluded that investments in social and economic infrastructure enhance private investment, output and productivity which may maximize the growth potential of economy. Samundram et al. (2009) analyzed whether Keynesian or Wagner s hypothesis hold for Malaysia by using ARDL bound testing approach for the period Empirical results indicated existence of cointegration between aggregate public spending including spending on education, agriculture, defense and development and GDP. Findings further indicated that long run causality is bi-directional in favor of both Keynes and Wagner s view with structural break for 1998 in case of GNP and expenditure on health and administration. Long run causality supports Wagner s hypothesis for all other expenditure categories. They concluded administration and health spending is mandatory while estimating long-run elasticity for relevant models. Afzal and Abbas (2010) examined existence of Wagner s hypothesis for economy of Pakistan by 7

10 cointegration methodology time span ranging from 1960 to Cointegration results indicated that aggregate expenditure, its subcomponents and output are not moving together in long-run. Results of causality elaborated unidirectional flow in favor of Wagner s law for total public spending, defence spending, interest payments and fiscal deficit. They suggested Wagner s hypothesis is valid for period , when fiscal deficit is included in case of public spending. Montiel (2010) examined relationship between public spending and output in Mexico. Cointegration and causality tests were applied for the period 1950 to Results suggested the presences of cointegration between public spending and output. Results of causality test suggested unidirectional causality which supports Wagner s hypothesis. He concluded fiscal policy do not act as a stabilizer of Mexico economy and stabilization function is not supported. Abdullah and Mamoor (2010) examined applicability of Wagner s hypothesis in Malaysia. ARLD and causality test were employed for the period 1970 to Results indicated long-run relationship between government expenditure and economic growth and significant in implementing developmental spending. Results also suggested that association between development expenditure and economic growth is validated in four out of five version of Wagner s law. They concluded Wagner s hypothesis is still applicable for Malaysia. Babatunde (2011) analyzed existence of Wagner s in Nigeria through employing ARDL bound testing approach for the period 1970 to Results explored that there is absence of cointegration between variables. Results of causality test indicated Wagner s hypothesis is not applicable to Malaysian economy. He concluded fiscal policy variables are major determinants of economic growth and public spending is an exogenous factor. Kumar et al. (2012) examined applicability of Wagner s hypothesis in New Zealand by using ARDL bounds test technique for time period 1960 to Results indicated public spending and output are moving together in long-run. Share of government spending in income may slightly effected by trade openness. Results of long run causality were of the view Wagner s hypothesis is valid for economy of New Zealand, but in short run Keynesian hypothesis hold true. Nworji et al. (2012) investigated association between government expenditure and economic growth in Nigeria by employing OLS multiple regression model for time span ranging from 1970 to Results showed capital and recurrent expenditure has negative and insignificant impact on national income. Output is positively and significantly affected by 8

11 expenditures on transfer payments, social and community services. The findings suggested both Wagner s and Keynesian hypothesis are applicable for Nigeria. They concluded that expenditures on social, economic and community services should be directed to productive economic activities to stimulate growth. Rauf et al. (2012) analyzed association between public spending and output in Pakistan. They employed ARL co-integration (2001) time span ranges from 1970 to The results indicated absence of long-run association between government expenditure and output. Causality results neither follow Wagner s nor Keynesian hypothesis for Pakistan indicating absence of any causal link between public spending and output. They concluded increase in public spending is not because of growth in output for Pakistan, this may occur due to other important factors. Magazzino (2012) analyzed association between aggregate public spending, its subcomponents and aggregate income for Italy. Cointegration and causality tests were applied for the period 1960 to Co-integration analysis revealed that expenditure on grants for production, passive interests as well as public investments and output are moving together in long-run. Causality results showed spending on dependent labor income in short-run and spending on passive interests in long run is line with validity of Wagner s hypothesis. On other hand, unidirectional causality is validated for Keynesian hypothesis in case of spending on grants for production, public investments and passive interests in long-run and grants for production in short-run. He concluded causality analysis were more support for Keynesian hypothesis than Wagner hypothesis. Oktayer (2013) examined association between public spending and economic growth for Turkey. ARDL bound testing approach was used time span ranging from 1950 to Cointegration and causality results with bivariate analysis neither indicated long-run relationship nor validity of Wagner s hypothesis. In the trivariate analysis by including inflation rate, revealed existence of cointegration, and also showed that Wagner s law hold. He concluded that inclusion of third variable (e.g. inflation) in system may change the whole scenario. Mahmoodi (2014) examined association between government expenditure and output for twenty Asian countries panel. Panel co-integration test was employed for the period 1970 to Empirical result indicated co-integration exists only for developing panel. For the case of advanced and newly industrialized countries, panel causality framework indicated that causality is bidirectional in short-run. Empirical evidence of developing panel showed in short run, unidirectional causality dictated Wagner s hypothesis, while in long-run bidirectional causality 9

12 exists. Barra et al. (2015) examined association between government expenditure and economic growth for Italy. Cointegration and causality test were employed time span ranging from 1951 to Results revealed the existence of cointegration between variables. Causality test suggested unidirectional causality for validity of Wagner s hypothesis. They concluded those public spendings which grow at a slower than economic growth can be reduced by establishment of Ministry of Finance. Muhammad et al. (2015) analyzed association between government expenditure and economic growth in Pakistan by employing cointegration approach for time span ranging from 1972 to Results showed public spending and output are not moving together in long-run. Causality results indicated no causal relations between public spending and national output which indicates no findings to support Keynesian and Wagner hypothesis in Pakistan. Growth rates are not achieved by expenditure, because these are not important tool for Pakistan. Lahirushan and Gunasekara (2015) investigated relationship between public spending and output for Asian countries. They used panel co-integration for the period 1973 to Results suggested absence of cointegration relationship between government expenditure and output in Asian region. Causality test results showed that both Keynesian and Wagner s hypothesis are valid. They concluded that prominent role is played by government to achieve growth in Asian region. Thabane and Lebina (2016) analyzed relationship between government expenditure and economic growth in Lesotho. Cointegration and causality tests were applied for the time span ranging from 1980 to Results indicated both government expenditure and economic growth are moving together in long-run. The causality results of short-run and long-run suggested Wagner s hypothesis is valid. They concluded that investment by government in physical infrastructure will enhance growth and improve fiscal sustainability as compared to recurrent expenditure. Therefore, government should reallocate its expenditure for investment in physical infrastructure. Literature represents mixed results of association between total government expenditure, its subcomponents and economic growth. Direction of causality for applicability of Wagner s and Keynesian hypothesis also give mix results. However, few of the studies found for Pakistan which investigated the relationship between aggregated as well as disaggregated expenditure and economic growth. There is less discussion on six basic versions of Wagner s law for Pakistan at aggregate and disaggregate level of expenditure. The reason behind testing government 10

13 expenditure at partial level is because, this analysis provides clear picture of the economy. Government can pay a special attention to a particular sector of the economy where serious actions are required. Therefore, it is necessary to inspect role of state towards development of nation. 3. Model, Methodology and Data 3.1 Model There are mainly two approaches regarding relationship between government expenditure and economic growth i.e. Wagner s law or Keynesian hypothesis. Wagner (1883), 19 th century German political economist formulated a principle called Law of the Increasing Extension of State Activity. He had not presented his ideas in form of a law, later on his views named as Wagner s hypothesis or Wagner s law (Henrekson, 1993; Halicioglu, 2003). The law argues government expenditure raises faster than economic growth and it is an endogenous policy variable. The government expenditure plays no role in generating economic growth, because government spending is a consequence rather than cause of economic growth. Therefore, causality flows from economic growth to public spending. Wagner s hypothesis cited that public activity level may exceed due to three reasons (Chang, 2002). First, expenditure on socioeconomic regulation, law and order situations may increase due to urbanization and agglomeration of population. Second, income elasticity of demand for goods provided by government such as health services, education, cultural activities and welfare expenditure is greater than unity. Third, government provides necessary funds in form of capital expenditures to fulfill technological needs of an industrialized economy which private organizations cannot provide. On the other hand, Keynes (1936) views relationship between government expenditure and economic growth is opposite to Wagner s hypothesis. The government expenditure is treated as exogenous policy instrument which is considered to influence economic growth and correct short-term as well as long-term cyclical fluctuations. Public spending is a cause rather than effect of economic growth, hence casualty runs from government expenditure to national income. Keynesian hypothesis treats demand as a prerequisite for growth. His analysis concludes economic performance may improve by demand management policies. However, inefficiencies of market failure can also improve by government intervention. 11

14 Various researchers (Musgrave, 1969; 1973) have challenged on Wagner s law, either law relates to absolute size of the government or relative size of the government in national economy, it is not clear. Musgrave & Musgrave (1984) comment that association between public and private goods is complementary instead of substitute in nature. Peacock and Wiseman (1967) brought up with displacement effect idea by using political theory to elaborate consequences of political events on government expenditure. Dutt and Ghosh (1997) commented Wagner was neither explicit in hypothesis formulation nor presented his law in mathematical form. Several mathematical specifications have been developed time to time by researchers to prove Wagner s or Keynesian view since 1960s for developed and developing countries. The following are widely acceptable functional forms: Model 1 The functional form of model 1 is presented by Peacock-Wiseman (1961). In this model, total government expenditure (GE) is a function of national output (GDP). If elasticity of GE with respect to GDP exceed unity, Wagner s hypothesis will supported and reveals public spending grows at faster rate than national output. The functional form of model 1 is as follow: GE = f (GDP) Where, GE is total government expenditure, and GDP is gross domestic product Model 2 The functional form of model 2 is proposed by Pryor (1968). In this model, government consumption expenditure (GCE) is a function of national output (GDP). Wagner s hypothesis is valid if elasticity of GCE with respect to GDP exceed unity. The functional form of model 2 is as follow: GCE = f (GDP) Where, GCE is government consumption expenditure, GDP is gross domestic product Model 3 The functional form of model 3 is related to Goffman (1968). In this model, the total government expenditure (GE) is as a function of per capita output (GDP/N). If elasticity of GE with respect 12

15 to GDP/N exceed unity, Wagner s hypothesis is supported. The functional form of model 3 is as follow: GE = f (GDP/N) Where, GE is total government expenditure, GDP is gross domestic product, and N is population Model 4 The functional form of model 4 is introduced by Michos (1975), a modified version of Gupta (1967). The applicability of Wagner s hypothesis requires elasticity of GE/N with respect to GDP/N exceed unity. The functional form of model 4 is as follow: (GE/N) = f (GDP/N) Where, GE is total government expenditure, GDP is gross domestic product, and N is population, Model 5 The functional form of model 5 developed by Mann (1980) which is the modified version of Peacock-Wiseman (1961). In this model, total government expenditure as a percentage of total output (GE/GDP) is a function of national output (GDP). If the elasticity of government share in total output with respect to national output exceed zero, Wagner s hypothesis is validate. The functional form of model 5 is as follow: (GE/GDP) = f (GDP) Where, GE is total government expenditure, and GDP is gross domestic product Model 6 Finally, the functional form of model 6 is firstly proposed by Musgrave (1969), later on modified by Ram (1986), Murthy (1993), Henrekson (1993) and Hseih and Lai (1994) respectively. In this model, the total government expenditures to national output (GE/GDP) is a function of per capita output (GDP/N). Wagner s hypothesis is valid if elasticity of GE/GDP with respect to GDP/N exceed zero. The functional form of model 6 is as follow: (GE/GDP) = f (GDP/N) Where, GE is total government expenditure, GDP is gross domestic product, and N is population Methodology 13

16 Econometric Model Wagner s (1883) hypothesis formulates public spending increases faster than economic growth. The share of GDP to public spending increases as an economy develops. Non-linear association between public spending and GDP leads specification of mathematical expressions of Wagner s law in exponential form. This study adopts six of functional forms of Wagner s law as discussed earlier. Mathematical formulation of six of Wagner s law depends upon exponential form can be written in equation form as: Model 1 The functional form of model 1 is presented by Peacock-Wiseman (1961). In this model, total government expenditure (GE) is a function of national outp ut (GDP). The equation of the model can be written as: ln (GE t ) = α o + α 1 ln (GDP t ) + u 1t (1) Where, GE t is total government expenditure, GDP t is gross domestic product, and u 1t is error term. The functional form which is presented by Peacock-Wiseman (1961) can be further extended by disaggregated expenditure as: ln (ECS t ) = β o + β 1 ln (GDP t ) + u 2t (1a) Where, ECS t is expenditure on current subsidies, GDP t is gross domestic product, and u 2t is error term. ln (ESEES t ) = γ o + γ 1 ln (GDP t )+ u 3t (1b) Where, ESEES t is expenditure on social, economic and education services, GDP t is gross domestic product, and u 3t is error term ln (ED t ) = δ o + δ 1 ln (GDP t )+ u 4t (1c) Where, ED t is expenditure on defence, GDP t is gross domestic product, u 4t is error term. ln (CE t ) = ρ o + ρ 1 ln (GDP t ) + u 5t (1d) Where, CE t is current expenditure, GDP t is gross domestic product, and u 5t is error term ln (DE) = ϴ o + ϴ 1 ln (GDP t ) + u 6t (1e) 14

17 Where, DE t is developmental expenditure, GDP t is gross domestic product, and u 6t is error term. Wagner s hypothesis is supported if α 1, β 1, γ 1, δ 1, ρ 1 and ϴ 1 exceed unity (>1) Model 2 The functional form of model 2 is proposed by Pryor (1968). In this model, government consumption expenditure (GCE) is based on national output (GDP). The equation of the model can be written as: ln (GCE t ) = α o + α 1 ln (GDP t ) + u 1t (2) Where, GCE t is government consumption expenditure, GDP t is gross domestic product, and u 1t is error term. Wagner s hypothesis is supported if α 1 exceed unity (>1) Model 3: The functional form of model 3 is related to Goffman (1968). In this model, the total government expenditure (GE) depends on per capita output (GDP/N). The equation of model can be written as: ln (GE t ) = α 0 + α 1 ln (GDP t /N t ) + u 1t (3) Where, GE t is total government expenditure, GDP t is gross domestic product, N t is population, and u 1t is error term. The functional form which is formulated by Goffman (1968) can be further extended by disaggregated expenditure as: ln (ECS t ) = β 0 + β 1 ln (GDP t /N t ) + u 2t (3a) Where, ECS t is expenditure on current subsidies, GDP t is gross domestic product, N t is population, and u 2t is error term. ln (ESEES t ) = γ 0 + γ 1 ln (GDP t /N t ) + u 3t (3b) Where, ESEES t is expenditure on social, economic and education services, GDP t is gross domestic product, N t is population, and u 3t is error term. ln (ED t ) = δ 0 + δ 1 ln (GDP t /N t ) + u 4t (3c) 15

18 Where, ED t is expenditure on defence, GDP t is gross domestic product, N t is population, and u 4t is error term. ln (CE t ) = ρ 0 + ρ 1 ln (GDP t /N t ) + u 5t (3d) Where, CE t is current expenditure, GDP t is gross domestic product, N t is population, and u 5t is error term. ln (DE t ) = ϴ 0 + ϴ 1 ln (GDP t /N t ) + u 6t (3e) Where, DE t is developmental expenditure, GDP t is gross domestic product, N t is population, and u 6t is error term. Wagner s hypothesis is supported if α 1, β 1, γ 1, δ 1, ρ 1 and ϴ 1 exceed unity (>1) Model 4: The functional form of model 4 is introduced by Michas (1975), a modified version of Gupta (1967). In this model, per capita total government expenditure (GE/N) is a function of per capita output (GDP/N). The equation of the model can be written as: ln (GE t /N t ) = α o + α 1 ln (GDP t /N t ) + u 1t (4) Where, GE t is total government expenditure, N t is population, GDP t is gross domestic product, and u 1t is error term. The functional form which is suggested by Michos (1975) can be further extended by disaggregated expenditure as: ln (ECS t /N t ) = β o + β 1 ln (GDP t /N t ) + u 2t (4a) Where, ECS t is expenditure on current subsidies, N t is population, GDP t is gross domestic product, and u 2t is error term. ln (ESEES t /N t ) = γ 0 + γ 1 ln (GDP t /N t ) + u 3t (4b) Where, ESEES t is expenditure on social, economic and education services, N t is population, GDP t is gross domestic product, and u 3t is error term. ln (ED t /N t ) = δ o + δ 1 ln (GDP t /N t ) + u 4t (4c) Where, ED t is expenditure on defence, N t is population, GDP t is gross domestic product, and u 4t is error term. 16

19 ln (CE t /N t ) = ρ o + ρ 1 ln (GDP t /N t ) + u 5t (4d) Where, CE t is current expenditure, N t is population, GDP t is gross domestic product, and u 5t is error term. ln (DE t /N t ) = ϴ o + ϴ 1 ln (GDP t /N t ) + u 6t (4e) Where, DE t is developmental expenditure, N t is population, GDP t is gross domestic product, and u 6t is error term. Wagner s hypothesis is supported if α 1, β 1, γ 1, δ 1, ρ 1 and ϴ 1 exceed unity (>1) Model 5 The functional form of model 5 developed by Mann (1980) which is modified version of Peacock-Wiseman (1961). In this model, total government expenditure as a percentage of total output (GE/GDP) is a function of national output (GDP). The equation of the model can be written as: ln (GE t /GDP t ) = α o + α 1 ln (GDP t ) + u 1t (5) Where, GE t is government expenditure, GDP t is gross domestic product, and u 1t is error term. The functional form which is presented by Mann (1980) can be further extended by disaggregated expenditure as: ln (ECS t /GDP t ) = β o + β 1 ln (GDP t ) + u 2t (5a) Where, ECS t is expenditure on current subsidies, GDP t is gross domestic product, and u 2t is error term. ln (ESEES t /GDP t ) = γ o + γ 1 ln (GDP t ) + u 3t (5b) Where, ESEES t is expenditure on social, economic and education services, GDP t is gross domestic product, and u 3t is error term. ln (ED t /GDP t ) = δ o + δ 1 ln (GDP t ) + u 4t (5c) Where, ED t is expenditure on defence, GDP t is gross domestic product, and u 4t is error term. ln (CE t /GDP t ) = ρ o + ρ 1 ln (GDP t ) + u 5t (5d) Where, CE t is current expenditure, GDP t is gross domestic product, and u 5t is error term. 17

20 ln (DE t /GDP t ) = ϴ o + ϴ 1 ln (GDP t ) + u 6t (5e) Where, DE t is developmental expenditure, GDP t is gross domestic product, and u 6t is error term. Wagner s hypothesis is supported if α 1, β 1, γ 1, δ 1, ρ 1 and ϴ 1 exceed zero (>0) Model 6 The functional form of model 6 is firstly proposed by Musgrave (1969), later on modified by Ram (1986), Murthy (1993), Henrekson (1993) and Hseih and Lai (1994) respectively. In this model, the total government expenditures to national output (GE/GDP) is a function of per capita output (GDP/N). The equation of the model can be written as: ln (GE t /GDP t ) = α o + α 1 ln (GDP t /N t ) + u 1t (6) Where, GE t is government expenditure, GDP t is gross domestic product, N t is population, and u 1t is error term. The functional form which is modified version of Musgrave (1969) can be further extended by disaggregated expenditure as: ln (ECS t /GDP t ) = β o + β 1 ln (GDP t /N t ) + u 2t (6a) Where, ECS t is expenditure on current subsidies, GDP t is gross domestic product, N t is population, and u 2t is error term. ln (ESEES t /GDP t ) = γ o + γ 1 ln (GDP t /N t ) + u 3t (6b) Where, ESEES t is expenditure on social, economic and education services, GDP t is gross domestic product, N t is population, and u 3t is error term. ln (ED t /GDP t ) = δ o + δ 1 ln (GDP t /N t ) + u 4t (6c) Where, ED t is expenditure on defence, GDP t is gross domestic product, N t is population, and u 4t is error term. ln (CE t /GDP t ) = ρ o + ρ 1 ln (GDP t /N t ) + u 5t (6d) Where, CE t is current expenditure, GDP t is gross domestic product, N t is population, and u 5t is error term. ln (DE t /GDP t ) = ϴ o + ϴ 1 ln (GDP t /N t ) + u 6t (6e) 18

21 Where, DE t is developmental expenditure, GDP t is gross domestic product, N t is population, and u 6t is error term. Wagner s hypothesis is supported if α 1, β 1, γ 1, δ 1, ρ 1 and ϴ 1 exceed zero (>0) Test of Stationarity Time series analysis are very sensitive with its non-stationary properties because if the series under consideration has unit root, it will mislead the results and conclusions as well. There are number of tests available to examine the presence of unit root. If series under consideration is stationary (mean and variance is constant), this implies that series do not follow unit root problem and integrated of order I(0). Dickey and Fuller (1979) have developed Dickey-Fuller unit root test in which they treat that error term (u t ) is uncorrelated. But in order to address the situation where error term is correlated, they have developed Augmented Dickey-Fuller unit root test in which they augmented their preceding test of unit root by adding lag of explained variable on right hand side. In ADF unit root test problem of serial correlation and hetroskedasticity is raised by adding lag of dependent variable on the right hand side. Phillips and Perron (1988) specially deals with this problem by employing nonparametric statistical methods without adding lag of explained variable Cointegration Test The application of Engle and Granger (1987) two step procedure requires I(1) order of integration of variables. This test explores whether series under consideration has long-run equilibrium relationship or not. There are two steps to examine cointegration which are given below as: Step 1 First step involves investigating long-run association between two variables under analysis through employing Ordinary Least Square (OLS) estimates. The general form of regression equation of OLS estimates is as follow: lng t = α 0 + α 1 lny t + ε t (7) Where G t represents total government expenditure or disaggregated expenditure, Y t is GDP and ε t represents error term. 19

22 The residuals obtained from regression equation (7) are tested for stationarity through employing ADF unit root tests. If estimated residuals (ε t ) are stationary at level, then variables are cointegrated and moving together in long-run. Step 2 After assessing long-run association between variables, we move towards 2 nd step. In this step short-run relationship is analyzed by estimating an ECM or dynamic model. If variables have possessed a long-run relationship, the residuals (ε t ) obtained from estimated regression equation can be employed to estimate the ECM (Enders, 2004). Granger representation theorem states that association between X and Y can be expressed as ECM, if these variables are moving together in long-run. The dynamic model involves in estimating the following equation: n n lng t = α 0 + i=1 α i lng t i + j=1 α j lny t j + αε t 1 + u t (8) Where, ε t is, the ECT is lagged estimated residual from equation (8). If variables are cointegrated, then the ECT should have negative and significant sign. The speed of adjustment of short-run equilibrium depends on the absolute value of the coefficient of error term Testing for Causality The previous section has provided information regarding the cointegration relationship between variables, but this does not allow us to draw conclusions about the direction of the causality between variables. Therefore, Granger causality test (1969) is applied to determine the existence of causality between variables. Engle and Granger (1987) states either unidirectional or bidirectional causality must exists for I(0) variables, if variables are integrated of order I(1) and cointegrated (Biswal et al. 1999). Granger causality test is investigated by the following equations: n n lng t = α 0 + i=1 α i lng t i + j=1 α j lny t j + u 1t (9) n n lny t = α 0 + i=1 α i lny t i + j=1 α j lng t j + u 2t (10) Where μ 1t and μ 2t are not correlated. The implications of causality test can emerge by following scenarios: i. If Σ α i 0 and Σ α j = 0 implying that causality runs from G to Y which interprets unidirectional relationship. 20

23 ii. iii. iv. If Σ α i = 0 and Σ α j 0, the causality runs from Y to G which is also of unidirectional in nature. If Σ α i 0 and Σ α j 0 implying that both coefficients on lagged values of regression equations are significant and diverge from zero. Therefore, causality is bidirectional from Y to G and G to Y in nature. If Σ α i = 0 and Σ α j = 0 implying both coefficients on lagged values of regression equations are not statistically different from zero. Therefore, causal link neither follow unidirectional nor bidirectional arrows between Y and G Data The annual time series data covering the period 1976 to 2015 is taken for Pakistan. The data is mainly collected from Handbook of Statistics on Pakistan Economy (2010) issued by State Bank of Pakistan and Pakistan Economic Survey (various issues). Detailed description of variables and their resources are given in Appendix A. 4. Results 4.1. Unit Root Tests Testing for unit root in order to avoid problem of spurious regression is the first step in time series analysis which can mislead the results. Augmented Dickey-Fuller (ADF) and Phillips- Perron (PP) test for unit root are extensively used to determine the order of integration. Schwarz Information Criterion (SC) and Akaike Information Criterion (AIC) are used for determination of appropriate lag order selection for each test. The results of unit root tests indicate all the variables are integrated of order I(1). Results of ADF and PP tests are reported in table

24 Variables Table 4.1: Results of ADF and PP Unit Root Tests Augmented Dickey Fuller Phillip Perron (PP) (ADF) Order of Integration Level 1 st Difference Level 1 st Difference ADF PP ln GE *** *** I(1) I(1) ln GDP *** *** I(1) I(1) ln ECS *** *** I(1) I(1) ln ESEES *** *** I(1) I(1) ln ED *** *** I(1) I(1) ln CE *** *** I(1) I(1) ln DE *** *** I(1) I(1) ln GCE *** *** I(1) I(1) ln(gdp/n) *** *** I(1) I(1) ln (GE/N) *** *** I(1) I(1) ln (ECS/N) *** *** I(1) I(1) ln(esees/n) *** *** I(1) I(1) ln (ED/N) *** *** I(1) I(1) ln (CE/N) *** *** I(1) I(1) ln (DE/N) *** *** I(1) I(1) ln (GE/GDP) *** *** I(1) I(1) ln (ECS/GDP) *** *** I(1) I(1) ln(esees/gdp) *** *** I(1) I(1) ln (ED/GDP) *** *** I(1) I(1) ln (CE/GDP) *** *** I(1) I(1) ln (DE/GDP) *** *** I(1) I(1) Note: ***, **, * shows significance at 1%, 5%, and 10% level respectively Results of Models Engle and Granger (1987) two step procedure is employed for long run association between variables, because all variables of models are integrated of order I(1). In first step, ADF test is 22

25 applied to check whether or not u t is stationary at level which means presence of long-run relationship. After estimating cointegration between variables, short-run adjustments are captured by Error Correction Model (ECM) in second step. Granger causality test is used for direction of causality between variables Results of Model 1 The results of cointegrating regression are reported in table 4.2. Results suggest that null hypothesis of no cointegration in model 1, model 1a, model 1c, model 1d and model 1e is accepted. For model 1b, null hypothesis of no cointegration do not accepted at 10% significance level which depicts presence of long-run relationship. In model 1, total government expenditure have proposed no long-run relationship with GDP. Results imply that increase is government expenditure with output is not due to increase in economic growth. This may because of other important factors i.e. urbanization, increase population growth, market exploitations and non-stability in political activates. These spending are not helpful to stimulate the pace of development. Ansari et al. (1997), Biswal et al. (1999), Iyare et al. (2004), Afzal and Abbas (2010) also found absence of cointegration association between total public spending and national income. Table 4.2: Results of Engle-Granger Cointegrating Regressions Model Coefficient of Dependent Adjusted Constant Explanatory Variable R-squared Variable CRDW 1 ln GE *** (0.1424) (0.0098) 1a ln ECS *** (0.8531) (0.0586) 1b ln ESEES *** (0.1978) (0.0136) 1c ln ED *** (0.2784) (0.0191) 1d ln CE *** (0.1968) (0.0135) 1e ln DE *** Calculated ADF Residuals Test ** * (0.3696) (0.0254) Note: The ln before variable name represents logarithm. ***, **, * shows significance at 1%, 5%, and 10% level respectively. CRDW stands for Cointegration Regression Durbin-Watson. The 1%, 5% and 10% critical value for the CDRW is 1.00, 0.78 and 0.69 respectively (Engle and Yoo, 1987). The 1%, 5% and 10% critical value for the ADF residual test is , and respectively (MacKinnon, 1991). Standard errors are in parenthesis. 23